ABSTRACT
The various feedforward and feedback control techniques discussed in the previous chapters have assumed considerable knowledge about the dynamics of the system being controlled. Multilayer feedforward neural networks and adaptive fuzzy sets are two structures that can accomplish such learning. This chapter provides a brief introduction to these methods and discusses their use in learning to back up a truck and trailer. From an abstract perspective, a dynamic system is a mathematical relation between a set of input variables and a set of output variables. The neural network can observe input-output pairs and adjust its parameters to approximate the poorly understood system. For a complex system, the mathematical description of the dynamic system may be unknown. This problem can be solved by training another neural network to mimic the dynamic system. More complex models with many parameters may also be used to describe human performance; the problem of parameter identification is more difficult and typically requires larger numbers of observations.
